mirrors/dragonfly
1
0
Fork 0
mirror of https://github.com/dragonflydb/dragonfly.git synced 2025-05-11 10:25:47 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

bug(sse2neon): fix sanitizer crash in _mm_loadu_si128 (#1037)

Browse source 
* bug(sse2neon): fix sanitizer crash in _mm_loadu_si128

Signed-off-by: adi_holden <adi@dragonflydb.io>

* remove static

Signed-off-by: adi_holden <adi@dragonflydb.io>

---------

Signed-off-by: adi_holden <adi@dragonflydb.io>
This commit is contained in:
adiholden 2023-04-10 10:56:58 +03:00 • committed by GitHub
parent 7887327fd4
commit 8600eacdc4
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 35 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

8
src/core/dash_internal.h
View file

@ -13,11 +13,7 @@
#include <cstring> #include <cstring>
#include <functional> #include <functional>
#if defined(__aarch64__) #include "core/sse_port.h"
#include "base/sse2neon.h"
#else
#include <emmintrin.h>
#endif
namespace dfly { namespace dfly {
namespace detail { namespace detail {
@ -891,7 +887,7 @@ uint32_t BucketBase<NUM_SLOTS, NUM_OVR>::CompareFP(uint8_t fp) const {
const __m128i key_data = _mm_set1_epi8(fp); const __m128i key_data = _mm_set1_epi8(fp);
// Loads 16 bytes of src into seg_data. // Loads 16 bytes of src into seg_data.
__m128i seg_data = _mm_loadu_si128(reinterpret_cast<const __m128i*>(finger_arr_.data())); __m128i seg_data = mm_loadu_si128(reinterpret_cast<const __m128i*>(finger_arr_.data()));
// compare 16-byte vectors seg_data and key_data, dst[i] := ( a[i] == b[i] ) ? 0xFF : 0. // compare 16-byte vectors seg_data and key_data, dst[i] := ( a[i] == b[i] ) ? 0xFF : 0.
__m128i rv_mask = _mm_cmpeq_epi8(seg_data, key_data); __m128i rv_mask = _mm_cmpeq_epi8(seg_data, key_data);

19
src/core/detail/bitpacking.cc
View file

@ -4,16 +4,11 @@
#include "src/core/detail/bitpacking.h" #include "src/core/detail/bitpacking.h"
#include "base/logging.h"
#if defined(__aarch64__)
#include "base/sse2neon.h"
#else
#include <emmintrin.h>
#include <tmmintrin.h>
#endif
#include <absl/base/internal/endian.h> #include <absl/base/internal/endian.h>
#include "base/logging.h"
#include "core/sse_port.h"
using namespace std; using namespace std;
namespace dfly { namespace dfly {
@ -43,7 +38,7 @@ static inline pair<const char*, uint8_t*> simd_variant1_pack(const char* ascii,
// Based on the question I asked here: https://stackoverflow.com/q/74831843/2280111 // Based on the question I asked here: https://stackoverflow.com/q/74831843/2280111
while (ascii <= end) { while (ascii <= end) {
val = _mm_loadu_si128(reinterpret_cast<const __m128i*>(ascii)); val = mm_loadu_si128(reinterpret_cast<const __m128i*>(ascii));
/* /*
x = ((x & 0x7F007F007F007F00) >> 1) | (x & 0x007F007F007F007F); x = ((x & 0x7F007F007F007F00) >> 1) | (x & 0x007F007F007F007F);
@ -81,7 +76,7 @@ static inline pair<const char*, uint8_t*> simd_variant2_pack(const char* ascii,
// Based on the question I asked here: https://stackoverflow.com/q/74831843/2280111 // Based on the question I asked here: https://stackoverflow.com/q/74831843/2280111
while (ascii <= end) { while (ascii <= end) {
val = _mm_loadu_si128(reinterpret_cast<const __m128i*>(ascii)); val = mm_loadu_si128(reinterpret_cast<const __m128i*>(ascii));
/* /*
x = ((x & 0x7F007F007F007F00) >> 1) | (x & 0x007F007F007F007F); x = ((x & 0x7F007F007F007F00) >> 1) | (x & 0x007F007F007F007F);
@ -114,7 +109,7 @@ bool validate_ascii_fast(const char* src, size_t len) {
__m128i has_error = _mm_setzero_si128(); __m128i has_error = _mm_setzero_si128();
if (len >= 16) { if (len >= 16) {
for (; i <= len - 16; i += 16) { for (; i <= len - 16; i += 16) {
__m128i current_bytes = _mm_loadu_si128((const __m128i*)(src + i)); __m128i current_bytes = mm_loadu_si128((const __m128i*)(src + i));
has_error = _mm_or_si128(has_error, current_bytes); has_error = _mm_or_si128(has_error, current_bytes);
} }
} }
@ -249,7 +244,7 @@ void ascii_unpack_simd(const uint8_t* bin, size_t ascii_len, char* ascii) {
const __m128i control = _mm_set_epi8(14, 13, 12, 11, 10, 9, 8, 7, -1, 6, 5, 4, 3, 2, 1, 0); const __m128i control = _mm_set_epi8(14, 13, 12, 11, 10, 9, 8, 7, -1, 6, 5, 4, 3, 2, 1, 0);
while (ascii < end) { while (ascii < end) {
val = _mm_loadu_si128(reinterpret_cast<const __m128i*>(bin)); val = mm_loadu_si128(reinterpret_cast<const __m128i*>(bin));
val = _mm_shuffle_epi8(val, control); val = _mm_shuffle_epi8(val, control);
rpart = _mm_and_si128(val, _mm_set1_epi64x(0x000000000FFFFFFF)); rpart = _mm_and_si128(val, _mm_set1_epi64x(0x000000000FFFFFFF));

26
src/core/sse_port.h Normal file
View file

@ -0,0 +1,26 @@
// Copyright 2023, DragonflyDB authors. All rights reserved.
// See LICENSE for licensing terms.
//
#pragma once
#if defined(__aarch64__)
#include "base/sse2neon.h"
#else
#include <emmintrin.h>
#include <tmmintrin.h>
#endif
namespace dfly {
inline __m128i mm_loadu_si128(const __m128i* ptr) {
#if defined(__aarch64__)
__m128i res;
memcpy(&res, ptr, sizeof(res));
return res;
// return vreinterpretq_m128i_s32(vld1q_s32((const int32_t *) p));
#else
return _mm_loadu_si128(ptr);
#endif
}
} // namespace dfly